Predictors of nurses’ response to informal carers’ interventions in medication safety and their collaboration with nurse colleagues and physicians: a cross-sectional factorial survey from Switzerland / Prädiktoren der Reaktion von Pflegefachpersonen auf Interventionen von betreuenden Angehörigen zur Medikationssicherheit und ihre Zusammenarbeit mit pflegerischem und ärztlichem Personal: eine faktorielle Querschnittstudie aus der Schweiz

In this cross-sectional, quantitative, factorial survey, randomised vignettes were evaluated by nurses. Each vignette presented a hypothetical case example in which a carer would approach the nurse regarding a claimed medication error. The vignettes contained a detailed description of the hypothetical situation, as well as the choice of words by which the carer would address the nurse. The participating nurses were asked to imagine that they were themselves the nurse who was being approached in the described way, and to provide their answers accordingly. For 1023 of the posed 1400 vignettes (73%), the participants indicated that they had already experienced a similar situation in the past, and 255 of the 285 participants (89%) did so for at least one vignette. In addition, individual characteristics of the participating nurses were collected as controls, giving the collected data a panel structure.

The factorial survey design has two distinct advantages whenever situations are tested in which multiple factors simultaneously influence an evaluation or a decision. Firstly, if one-dimensional approaches were used instead, respondents may find it difficult to isolate the marginal effect of a single parameter, when, in fact, several parameters are relevant to their decision-making. It would, therefore, likely be more cumbersome and unusual compared to their everyday work than just reacting to a situation they encounter. The factorial design avoids this drawback. Secondly, the factorial design is capable of handling a relatively large number of parameters (and their combinations) while posing comparatively few questions (vignettes) per participant. For a more extensive description of the factorial vignette design, see, for example, Auspurg and Hinz (2015).

We employed a research design in line with Davis et al. (2014), however departed from it in the following ways: (1) The participants were not only asked about how appropriate they perceived interventions by carers to be (i.e. about their inner attitude), but also about their verbal response and whom they would verify the medication with. (2) To simulate the situation in which a nurse makes her/his decisions, the information of whether or not a medication error was present (i.e. the result of the following verification) was not included in the vignettes. (3) A set of individual attributes of the nurses were collected to serve as controls. (4) The vignette parameter of the intervening person claiming that she/he had professional expertise in the field (healthcare) was included. (5) The appropriateness of the interventions was evaluated on a discrete rating scale instead of a Likert scale (see ‘Questionnaire’).

Swissethics, the governing body of Swiss ethics committees on research involving humans, decided based on the research proposal that no ethical approval was required. Participation in the survey was voluntary and anonymous, and no data on the health of the participants or their patients were collected.

The data were collected by two-stage cluster sampling. A standardised online questionnaire was sent to all members of the Swiss Association of Nurses’ regional chapter of the three cantons of Zurich, Glarus and Schaffhausen (approx. 3000 members) for voluntary participation (Schweizerischer Berufsverband der Pflegefachfrauen und Pflegefachmänner [SBK/ASI]). The invitations and the links to the questionnaire were delivered to the members’ personal e-mail addresses on 2nd October 2019, a reminder was delivered after one month, and participation was open until 2nd December 2019. In 366 cases (12%), the link to the questionnaire was used, and 285 participants (9%) completed the questionnaire. Average completion time was 13 minutes (minimum 5; median 10; maximum 53). A total of 1400 vignettes were evaluated by 285 nurses (with 266 nurses answering all five posed vignettes, 16 nurses answering four and three nurses answering two). The participants’ characteristics are listed in Table 1.

The questions were posed with answer options in multiple-choice or multiple-answer form, thereby ensuring replicability and facilitating comparisons. Exceptions to this were dates, time durations and percentages. Furthermore, five of the multiple-choice questions included the answer option ‘other, namely’, which allowed for specification via text input to capture any otherwise potentially missing dimensions. The questionnaire consisted of five vignettes per participant and a set of questions on individual characteristics.

The first part of each vignette described the specific situation along five subsequent dimensions, each of which contained one of the following phrases. Therein, ‘you’ meant the participating nurse who would be faced with the hypothetical situation: (1) The presence or absence of the patient was introduced by either ‘Someone is sitting at the bed next to the patient and addresses you as you enter the room’ or ‘You are being approached on in the corridor’ (2–4) Age (25, 50 or 75 years) and gender (male or female) of the intervening carer, as well as her/his relationship with the patient (daughter/son, partner, father/mother or the ambiguity of the relationship), were stated as ‘It is Ms Müller, who is about 25 years of age and the daughter of your patient’, ‘It is Mr Schmid, who is about 50 years age and the partner of your patient’, ‘It is a woman of about 75 years of age, who apparently has some kind of a relationship with your patient (however you do not know whether she is a relative or an acquaintance)’, etc. (5) The health state of the patient was indicated by ‘The patient has been severely ill for several days’, ‘The patient had been hospitalised as an emergency yesterday’ or ‘The patient will leave the hospital tomorrow’. In the second part, each vignette covered three dimensions of direct speech by the intervening carer towards the nurse: (6) The verbal address was displayed as ‘Excuse me. Could you please check the tablets?’, ‘There is something wrong with the tablets’ or ‘Hello you! You have made a mistake here’. (7) The specific input regarding the (supposed) medication error was ‘So far, no tablets have been given all day. However, one should be taken every day, right?’, ‘So far, two tablets have been regularly given around noon. However, today there are three of them’, ‘Today the tablet is different from that of yesterday’ or ‘So far, drops have been regularly given around noon. Today, there are tablets’. (8) Some of the vignettes contained the additional phrase ‘I am also a professional in the field myself’.

Each nurse answered the same four questions for each vignette, the first three of which directly addressed the research questions: (1) ‘How appropriate do you find this behaviour?’ – Answers were given on a discrete rating scale ranging from 0 to 10, with only the extremes having an additional label: ‘0: entirely inappropriate’, ‘10: entirely appropriate’. (2) ‘Would you address the person about how she/he should behave in the future?’ – with the multiple-choice options ‘Yes, encourage them to approach me again if anything is unclear’ (encouraging option), ‘No, not specifically’ (neutral option), ‘Yes, signal that they should not interfere’ (rejecting option). (3) ‘What would you do in this situation?’ – with multiple-choice options given as ‘Check again together with the physician in charge whether there is a mistake’, ‘Check again together with another nurse whether there is a mistake’, ‘Check again myself whether there is a mistake’ and ‘Don’t check again whether there is a mistake’. (4) ‘Have you experienced a similar situation before yourself?’ – ‘yes’ or ‘no’. The answer to this last question was used as a predictor rather than an outcome (see ‘Data analysis’).

The set of questions on individual attributes covered the share of a participant's working time regularly spent in direct contact with carers, her/his role as a double-duty carer, education, work experience, hierarchical position, healthcare sector, specialised field, gender, age, the country in which she/he had passed the majority of her/his education and the duration of her/his membership in SBK/ASI.

For each research question, the marginal effects of the situational vignette parameters on the respective outcome of interest were estimated by a regression model with intercepts for individuals. Robust standard errors were used during systematic elimination of statistically insignificant predictors. For any tests of hypotheses, a type-one error probability (p) <0.05 was considered statistically significant. Serial effects of the order of vignettes, that is, whether earlier vignettes were answered systematically differently from the later ones, were tested by logistic regression without any significant evidence found. The shape of the relationship of interval-scaled predictors (age, working time spent with carers) with the respective outcome was tested in order to ensure model fit.

Perceived appropriateness was modelled by a fractional logit regression as proposed by Papke and Wooldridge (1996), with individual intercepts as in Wagner (2003), and applied to discrete rating scales by Studer and Winkelmann (2017) (Model 1). In contrast to Likert scales, the methodology by Studer and Winkelmann (2017) satisfies the assumptions of fractional regression when modelling (quasi-)metric ratings (see ‘Questionnaire’). As noted by Papke and Wooldridge (2008), a so-called ‘potential incidental parameters problem’ is inherent to the conditional logit model with intercepts in the case of a non-dichotomous outcome. To investigate this, perceived appropriateness was in addition modelled by a fractional probit panel regression with fixed effects, which, although possibly less popular in the literature, enables consistent estimation of marginal effects in the mentioned case (Model 2). The marginal effects of the individuals’ attributes were estimated by a pooled fractional logit regression (Model 3), as well as an alternative linear ordinary least squares (OLS) regression (Model 4). 1390 vignettes from 283 nurses were included in the latter two models since two nurses did not complete all questions on the individual attributes. While an analysis of residuals showed that the OLS model violated the assumptions of homoscedasticity and normality, but not the assumption of unbiasedness, the marginal effects did not differ substantially from the logit model. The results of the logit and probit regression models were computed as average marginal effects (AME; see e.g. Jann, 2013). The interpretation of an AME of, say, −0.7 on the discrete outcome scale of perceived appropriateness, therefore, reads as, for example, ‘on average over all vignettes, acceptance is 0.7 points lower if the relationship between the intervening person and the patient is/were unknown to the nurse, ceteris paribus’.

The verbal reply by nurses was modelled by an ordered logit panel regression with fixed effects, thereby employing the ‘blow up and cluster’ (BUC)-estimator according to Baetschmann et al. (2015), and using the algorithm provided by Dickerson et al. (2011) (Model 5). Pooled ordered logit regression estimated the marginal effects of the individuals’ attributes (Model 6). The underlying proportional odds assumption was assessed by the according multinomial models. The results were calculated as usual odds factors. Given proportional odds, an odds factor of, say, 0.07 is to be interpreted as, for example, ‘if the relationship between the carer and the patient is unknown to the nurse, the odds of the verbal reply to be at a more affirming level (encouraging > neutral > rejecting) multiply by 0.07, ceteris paribus’. This means that, if the odds of an encouraging reply were projected to be, say, 1.71, considering all other factors besides relationship, then the relationship between the intervening person and the patient being unknown to the nurse would reduce the projected odds of an encouraging reply from 1.71 to 0.12 (or e.g. its probability from 63.1% to 10.7%). Model 5 was approximated for 423 vignettes from 86 individuals with variation in the outcome, while individuals without variation chose predominantly the encouraging option (96.1%), and Model 6 was approximated for 1315 vignettes from 263 individuals.

Acts of collaboration in the verification process of the medication were modelled by a multinomial logit panel regression with fixed effects, as described by d’Haultfoeuille and Iaria (2016) and Pforr (2014) (Model 7). A pooled model as described by Greene (2012) assessed the marginal effects of the individuals’ attributes (Model 8). Verification by oneself (without other nurses/physicians) was chosen as the reference category of the nominal outcome, and the results were computed as specific odds factors to the alternatives with a nurse colleague or a physician. For example, an odds factor of 0.3 for old age specific to the outcome involving a physician is to be interpreted as ‘if the intervening person is of old age, the odds of the nurse verifying the medication together with a physician instead of by her-/himself multiply by 0.3, ceteris paribus’. Hence, if the odds of verification together with a physician were projected as, say, 0.23, considering all other factors, the intervening person being of old age would reduce these odds to 0.07 (or e.g. the probability from 18.6% to 6.5%). Model 7 was approximated for 502 vignettes from 104 individuals with variation in the outcome, and Model 8 model was approximated for 1318 vignettes from 267 individuals. The option ‘Don’t check again whether there is a mistake’ was chosen for 11 vignettes (0.8%), which did not enter the multinomial model.

Perceived appropriateness of vignettes had a mean of 7.67, a median of 8.0 and a median absolute deviation (MAD) of 2.97 on the scale of 0 (labelled ‘completely inappropriate’) to 10 (labelled ‘completely appropriate’). Each of the possible outcome values on the discrete rating scale (0 through 10) was chosen as an answer multiple times, with 10 being the most frequent (554 vignettes, 40%), 1 being the least frequent (24 vignettes, 2%) and with 34 mentions (2%) of the value of 0. The AME of Models 1 and 2 are listed in Table 2. Perceived appropriateness was significantly higher (p < 0.001) among nurses who had already encountered a similar situation to the one evaluated in the past, with an AME of 1.01 (logit model) and 0.92 (probit model). By contrast, perceived appropriateness was as much lower, if the verbal intervention was accusatory (‘Hello you! You have made a mistake here…’), with AME_logit = −1.86 and AME_probit = −1.80 (p < 0.001). This was the strongest negative marginal effect in the intercept/panel model. Furthermore, perceived appropriateness was lower if the verbal intervention was not personally accusatory but stating a mistake (‘There is something wrong with the tablets…’, p < 0.001, AME_logit = −0.39, AME_probit = −0.45), if the relationship between the intervening person and the patient was unclear (p < 0.001, AME_logit = −0.75, AME_probit = −0.70) and if the intervening person claimed to have professional expertise in the field her-/himself (p < 0.05, AME_logit = −0.18, AME_probit = −0.17). Hence, the probit model confirmed the results of the logit model, as shown by similar results.

Several individual attributes of the healthcare staff had a significant effect on perceived appropriateness, as shown in Table 3 (Models 3 and 4). It was lower among young nurses up to the age of 30 years (p < 0.01, AME_logit = −0.69, AME_OLS = −0.71) and also among staff with more than 40 years of professional experience in patient care (p < 0.01, AME_logit = −1.29, AME_OLS = −1.38). This means that among the upper quantiles of age, it is the long-term job experience that has an impact on acceptance rather than the higher age itself (the latter of which was also tested, leading to no significant results). Nurses with a Bachelor's degree (BSc) (p < 0.05, AME_logit = 0.43, AME_OLS = 0.42) and nurses who had completed a professional development education (p < 0.001, AME_logit = 0.81, AME_OLS = 0.82) displayed higher perceived appropriateness. Perceived appropriateness was higher if a nurse had a position as a care expert (‘Pflegeexpertise’) (p < 0.001, AME_logit = 1.19, AME_OLS = 1.28), and it was lower in the sectors of home care (p < 0.01, AME_logit = −0.49, AME_OLS = −0.48), nursing homes (p < 0.01, AME_logit = −0.62, AME_OLS = −0.63), healthcare consulting (p < 0.05, AME_logit = −1.90, AME_OLS = −1.88) and among freelancing nurses (p < 0.001, AME_logit = −2.61, AME_OLS = −2.54). Also, perceived appropriateness was lower in the field of somatic care for adults (compared to paediatrics, geriatrics, psychiatry and others) (p < 0.001, AME_logit = −1.23, AME_OLS = −1.14). Furthermore, nurses who had passed the majority of their education in Germany displayed higher perceived appropriateness compared to those who passed their education in Switzerland or the Netherlands (p < 0.001, AME_logit = 1.18, AME_OLS = 1.11), with the latter two groups not being significantly different from each other regarding the outcome. Members from other countries displayed lower acceptance than the Swiss and Dutch on average, with a difference that was, however, not statistically significant (p = 0.065, AME_logit = −1.03, AME_OLS = −1.07). Accordingly, for the latter effect, coincidence during sampling cannot be excluded as a cause with sufficient probability. Finally, nurses who indicated that they had formerly been double-duty carers displayed lower perceived appropriateness (p < 0.01, AME_logit = −0.39, AME_OLS = −0.38) than their nurse colleagues who had never been double-duty carers. As these results show, the AME of point estimates were robust across the logit and the OLS model.

For a proportion of 63% of the vignettes of Model 5 the encouraging option was chosen (30% neutral, 7% rejecting). Figure 1 illustrates both the relative frequency (probability) and the odds of each reply option within this sample, which are related by:

Figure 1

Table 4 presents the coefficients of the ordered logit panel regression on the log-odds scale, as well as the according odds factors. Similar to the results on perceived appropriateness, a more affirming reply (encouraging instead of neutral; neutral instead of rejecting) was significantly more likely (p < 0.01) if a respective nurse had already encountered a similar situation in the past, with an odds factor of 3.40. Applied to, for example, the odds of an encouraging reply of 1.71 (63.1%), this increased the odds to 5.83 (85.4%) for a respective individual. Furthermore, the more affirming reply levels were significantly less likely if the relationship between the intervening person and the patient was unclear (p < 0.001, odds factor = 0.07), if the verbal intervention occurred in the presence of the patient (in the patient's room) instead of in the absence of the patient (in the corridor) (p < 0.001, odds factor = 0.17), if the verbal intervention was accusatory (‘Hello you! You have made a mistake here…’) (p < 0.01, odds factor = 0.35) and if the patient had just been hospitalised as an emergency the day before (p < 0.05, odds factor = 0.47).

According to Model 6 (see Table 4), nurses up to the age of 30 years were more likely to provide a more affirming reply (p < 0.001, odds factor = 2.76), and so were those with a BSc or Master's (MSc) degree (p < 0.05, odds factor = 1.95), those working in the sector of home care (p < 0.01, odds factor = 1.90) and those who had passed the majority of their education in Germany (p < 0.01, odds factor = 9.54). A lower degree of professional experience in patient care up to 10 years (p < 0.05, odds factor = 0.59) and freelancing (p < 0.001, odds factor = 0.26) were associated with a lower likelihood of an affirming reply. Also, nurses who indicated that they had been double-duty carers, but had given it up by the time of the survey (p < 0.01, odds factor = 0.57) were less likely to provide an affirming reply.

The participants most often decided that they would verify the medication by themselves (for 53% of the vignettes), while for 31% of the vignettes they chose joint verification with a nurse colleague and for 16% verification with a physician (Model 7). The relative frequencies (probabilities) and odds of these answers are illustrated in Figure 2.

Figure 2

As shown in Table 5, involving a physician (p < 0.01, odds factor = 0.26), as well as including a nurse colleague in the verification process (p < 0.01, odds factor = 0.28) were less likely if the intervention occurred in the presence of the patient (in the patient's room). Also, the inclusion of a physician was less likely if the intervening person was elderly (75 years instead of 50 or 25 years) (p < 0.01, odds factor = 0.30). By contrast, the involvement of a nurse colleague was more likely if the intervening person claimed to have professional expertise in the field her-/himself (p < 0.05, odds factor = 1.67).

Among the individuals’ attributes that increased the likelihood of involving another person in the verification process (nurse colleague or physician) was the proportion of a nurse's working time regularly spent in direct contact with carers (Model 8). This proportion ranged between 0% (10 vignettes) and 100% (43 vignettes) with a lower quartile of 10%, a median of 20% and an upper quartile of 50%. The pooled regression revealed that an increase of 10 percentage points on this scale led to a marginal odds factor on the likelihood of involving a physician of 1.18 and of involving a peer of 1.17 (p < 0.001 in both cases). This means that if an individual had a proportion of, for example, 50% of her/his time spent with carers instead of 10%, her/his odds of involving a physician multiplied by 1.94 (=1.18⁴) and her/his odds of involving a peer instead multiplied by 1.87 (=1.17⁴). Similarly, higher age of a nurse led to a marginal odds factor per 10-year difference of 1.74 for including a physician (p < 0.001) and of 1.29 for the likelihood of including a peer (p < 0.01). Hence, if an individual was 50 years instead of 30 years old, her/his odds of involving a physician multiplied by 3.03 (=1.74²) and her/his odds of involving a peer instead multiplied by 1.66 (=1.29²). In addition, nurses who at the time of the survey were double-duty carers were more likely to include a peer in the verification process (p < 0.05, odds factor = 1.47). Conversely, involving a nurse colleague was less likely within the field of paediatrics (p < 0.05, odds factor = 0.43).

It should be noted that all the effects as estimated by the regression models presented here are to be interpreted ceteris paribus.

The previous professional experience of nurses proved important, as nurses who had already encountered a similar situation with carers in the past were significantly more accepting and more likely to answer encouragingly. Also, if nurses regularly spent a higher percentage of their working time in direct contact with carers, they were more likely to involve another nurse or a physician to verify the medication. Hence, specific working experience with carers seems to contribute to overcoming nurse-carer barriers, rather than mere general working experience in nursing irrespective of carers. Reflective practice contributes to honouring carers’ involvement in healthcare practice (Frampton et al., 2017), and family and patient-centred care ‘… includes a group of dynamic, positively reinforcing actions rather than a linear set of activities’ (Balik et al., 2011, p. 3). The latter statement is particularly relevant because safety issues can arise at any time in a dynamic healthcare environment. Hence, a mental ‘tick box’ does not serve the purpose of a safety culture. The results further point towards a known concept of carers’ needs for three main attitudes and skills of healthcare professionals: active listening, communication competence and empathy (Brügger et al., 2020). However, the results also emphasise the importance of the attitude brought forward by the carers in the verbal exchange, which turned out as the strongest driver of how the nurses perceived their involvement: If the carer initially accused the nurse of having made a mistake (i.e. by saying ‘Hello you! You have made a mistake here’), the intervention as a whole was perceived as much less appropriate than if the initial address was polite (‘Excuse me. Could you please check the tablets?’), and it caused an encouraging reply by the nurse to be much less likely. If the carer instead opened with ‘There is something wrong with the tablets’, the intervention was still perceived as less appropriate than if the address was polite, but to a much smaller extent. Perceived appropriateness was also significantly lower if the relationship between the intervening carer and the patient was unclear, which further made an encouraging reply by the nurse to be considerably less likely. In addition, if the carer highlighted to have professional expertise in the field without being asked about it, appropriateness was also lower, indicating that double-duty carers are perceived as interfering in the professional sphere (Jähnke et al., 2017). All these effects, as are all other estimates, are to be interpreted ceteris paribus, that is, irrespective of the other parameters of the situation.

An encouraging reply by the nurse, as well as involving a co-worker in the verification of the medication, was shown to be less likely if the patient was present during the exchange, which may appear somewhat surprising, since the so-called trialogue of patients, carers and professionals is considered the standard for a supportive working relationship at least in psychiatric care (Burr et al., 2018). It may be argued, however, that hospital nurses outside psychiatry may see their authority more challenged if their work is questioned in the presence of a patient instead of a private one-to-one conversation with the carer only.

Perceived appropriateness was higher if a nurse had a position as a nursing expert (‘Pflegeexpertise’), for example, as an advanced practice nurse, or if a nurse had completed a continuing education post-diploma education or a BSc degree. Similarly, the likelihood of a more affirming answer was higher if the nurse had a BSc or MSc degree. These results are in line with several findings from previous international studies, some including Switzerland, indicating that a BSc degree in nursing increases patient safety (Aiken et al., 2014; Djukic et al., 2019).

Relying on interprofessional collaboration can be an important measure when confronted with medication safety-relevant involvement by carers. In less than every sixth tested case in this study, the nurses indicated that they would include a physician. Interestingly, three of the main results regarding collaboration were very similar between the inclusion of a physician or a nurse colleague: The likelihood of involving another nurse or a physician in the verification of the medication increased along with the nurses’ age. As mentioned above, if the nurses regularly spent a higher percentage of their working time in direct contact with carers, she/he was more likely to involve both a nurse and a physician, and if the patient was present during the intervention, involving a nurse or a physician was less likely. Hence, involving coworkers either from medicine or nursing seems to be rather symmetric. Only when carers were elders (75 years instead of 50 or 25 years), and when the carer claimed to have professional expertise, the participants referred differently to the two groups of co-workers.

In factorial survey designs, the vignettes are of hypothetical nature and transferability of the results to ‘real life’ needs to be addressed. While, typically, both transferability and non-transferability remain unprovable hypotheses, a useful indicator may lie in whether participants tended to find the scenarios relatable or realistic. Also, while 89% of the participants claimed that they had experienced a similar situation in the past at least once may serve as a confirmatory indication, it is no proof. (It needs to be noted that a specific participant not having experienced a similar situation in the past does not automatically imply that she/he found the scenario itself unrelatable or unrealistic.)

Membership in the SBK/ASI is not compulsory. Therefore, any systematic selection effects of nurses into the SBK/ASI relevant to this study, or any systematic participation in the survey within the SBK/ASI, cannot be ruled out. However, since the study was designed to estimate the effects of the determinants of the responses by nurses, instead of merely the prevalence of their responses, in order for the results to be affected by a selection bias regarding any given attribute, there needs to be a significant interaction between the effects of the determinants and the attribute.

A larger sample size, although not necessarily affecting the unbiasedness of estimators, could decrease standard errors and increase the probability of significant results of inferential statistical tests.